fynance 1.6.0

Python and Cython scripts of machine learning, econometrics and statistical features for financial analysis

Fynance

---

Python and Cython package providing machine learning, econometric and statistical tools for financial analysis and backtesting of trading strategies.

Installation

pip install fynance

From source:

git clone https://github.com/ArthurBernard/Fynance.git
cd Fynance
pip install -e ".[dev]"
python setup.py build_ext --inplace

Subpackages

Algorithms fynance.algorithms
Portfolio allocation methods (ERC, HRP, IVP, MDP, MVP), walk-forward wrappers, and position sizing (fractional Kelly, volatility targeting, transaction costs).

Backtest fynance.backtest
Profit-and-loss plotting and performance measurement.

Estimator fynance.estimator
Cython ARMA / GARCH parameter estimation.

Features fynance.features
Kalman filter, technical indicators (Bollinger, RSI, MACD, ROC, realized volatility, rolling skew/kurtosis/autocorr, …), statistical momentums (SMA, EMA, WMA, …), metrics (Sharpe, Sortino, Calmar, drawdown, tail ratio, …), scaling (incl. rolling rank), feature-engineering tools (multi-resolution, Granger causality), and market-regime detection.

Models fynance.models
Econometric models (MA, ARMA, ARMA-GARCH), neural networks with PyTorch (MLP, RNN, GRU, LSTM, MultiHeadAttention, TCN, Transformer), a direction+magnitude stacking ensemble, differentiable loss functions (Sharpe, Sortino, Calmar, Omega, directional, hybrid), robust-training utilities (purged CV, early stopping, sample weighting), and walk-forward rolling evaluation.

Quick start

import numpy as np
import fynance as fy

# Sharpe ratio
returns = np.random.randn(252) * 0.01
print(fy.sharpe(returns))

# ERC portfolio allocation
cov = np.cov(np.random.randn(5, 252))
weights = fy.ERC(cov)
print(weights)

Rolling walk-forward training with a neural network:

import torch
import torch.nn as nn
from fynance.models.rolling import RollMultiLayerPerceptron

model = RollMultiLayerPerceptron(X, y, layers=[64, 32])
model.set_optimizer(nn.MSELoss, torch.optim.Adam, lr=1e-3)
model(train_period=252, test_period=21, roll_period=21)  # walk-forward windows
for eval_set, test_set in model:   # each step trains on the past, tests the next
    model._training()

See Notebooks/pytorch_examples.ipynb for a runnable tour (metrics, allocation, MLP/TCN/Transformer with custom losses, walk-forward CV).

Links

• PyPI: https://pypi.org/project/fynance/
• Documentation: https://fynance.readthedocs.io/en/latest/
• Source: https://github.com/ArthurBernard/Fynance
• Changelog: https://github.com/ArthurBernard/Fynance/blob/master/CHANGELOG.md